Ollama: Proxy Transfers and API Resilience
Merged fixes harden registry downloads behind proxies and make the Responses API tolerant when switching to non-thinking models. Open proposals target raw output tokens, large-number precision, and model lookup races, alongside cloud usage docs and a wave of community integrations.
Duration: PT2M33S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-10-17T13:00:57Z
- Audio duration: PT2M33S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Hello, it's October 17th, 2026, and this is your Ollama developer briefing.
The big pattern this cycle is resilience at the edges — downloads behind corporate proxies, switching between thinking and non-thinking models, and small server correctness fixes that prevent silent failures.
First, registry transfers. PR 19065, now merged, changes how redirect downloads behave behind a proxy. Instead of resolving the redirect target locally, Ollama defers that DNS lookup to the proxy, while keeping the stricter resolve-and-pin path when no proxy is set. It also trusts default registry and CDN host names…
Second, API compatibility. PR 19010, also merged, fixes a friction point in the OpenAI Responses interface. If a client moved from a thinking model to a non-thinking model like Llama 3.2, the saved high effort setting carried over and the turn was rejected as unsupported. Ollama now applies its non-thinking fallback…
Finally, correctness and ecosystem. Open proposals would preserve parser tokens in raw generate output for models like Gemma 4 in PR 19066, preserve 64-bit precision for large seeds that JavaScript-style numbers would round in PR 19081, and remove a lock contention flake in…
What'…
Nearby episodes from Ollama
- Model Compatibility Catch-Up and Windows Reliability
- Responses Compatibility and Clearer Errors
- Decision Models Go Usable, MLX Speeds Up on CUDA
- Decision Scoring Matures as Inference Hardens
- Native Decision Models and Precision Fixes
- Weekly Recap - Decision Models Go Native, Embeddings Go Multimodal
- GPU Discovery Fallback and Docs Cleanup
- OpenAI Parity and Reasoning Fixes